Combined Central and Subspace Clustering for Computer Vision

Combined Central and Subspace Clustering for Computer Vision Application Le Lu, Rene Vidal John Hopkins University (担当：猪口)

Introduction • Central Clustering – クラスタの中心の周辺にデータが分布 – Application • Image segmentation, – K-means，EM • Subspace Clustering – 部分空間にデータが分布 – Application • Motion segmentation, face clustering with varying illumination, temporal video segmentation – K-subspace, Generalized PCA

Algorithm GPCA

• Computing the membership • Computing the cluster centers – を で偏微分して， を掛けると • Computing the normal vectors – 上と同様 が使えて

Experiments (Simulated Data) KK KM KK MP GK JC • • • KM MP KK GK JC K-mean → 6つのクラスタを 2つの平面に分ける MPPC (Mixture of probabilistic PCA )→ 6つのクラスタを 2つの平面に分ける K-subspace→それぞれのSubspaceでK-means GPCA→それぞれのSubspaceでK-means 提案手法 KM GK MP JC

Experiments (Illumination) • 4 subjects (10 subjectのうち) • 4 poses × 64 illuminations • 240 × 320 pixels

• GPCA+K-means • Subject 5とSubject 6の交わりをSubject 5にクラスタリング

Experiments (Video) • Video sequence → several video shots • Each video contains 4 shots